Automated inspection is a technique for measuring the integrity of an object by collecting an image of it and comparing that image to a reference (e.g., comparing a die to a data-base for photolithographic masks), to another part of that object (such as die-to-die inspection for semiconductor wafers), or to a reference image (die-to-“golden image”). Disadvantageously, when conducting high-resolution inspection of large semiconductor substrates, the FOV of the imaging system cannot cover the entire substrate to be inspected, so the substrate must be moved or “stepped” across the FOV, thereby increasing inspection time. To increase throughput, some conventional automated inspection tools continuously scan the substrate in one direction while optically imaging an orthogonal one-dimensional optical FOV. Once the substrate is traversed in the scanning direction, it is typically moved in the other (cross-scan) direction by a distance of one FOV, and then its path is retraced, creating a serpentine motion path.
Other optical imaging systems for inspecting semiconductor substrates utilize a “spot grid array” to achieve high throughput. In these systems, an imager typically includes a two-dimensional and periodic array of lenses, each lens imaging a spot in an object plane, such as a substrate to be inspected, upon an image plane to image a two-dimensional and periodic array of spots from the object plane upon the image plane. A sensor, such as a CCD, is provided in a conjugate image plane with a two-dimensional and periodic array of readout elements, each collecting the signal from a spot in the object plane. A mechanical system moves the substrate such that as the substrate is moved across the spot array in the scan direction (the y-direction) the spots trace a path which leaves no gaps in the mechanical cross-scan direction (the x-direction). Thus, imaging of very large FOVs is accomplished by employing an array of optical elements each having a minimal FOV, rather than complex large-FOV optics. Optical imaging devices utilizing a spot grid array are described in U.S. Pat. No. 6,248,988 to Krantz, U.S. Pat. No. 6,133,986 to Johnson, U.S. Pat. No. 5,659,420 to Wakai, and U.S. Pat. No. 6,043,932 to Kusnose.
These and other previous implementations of spot-grid array concepts suffer from several limitations. To achieve the very high data-rates required for high-end inspection with all-mechanical stage scanning, a large array is required. However, some major problems prevent the use of prior art technologies for large arrays, such as relatively limited focus capabilities, imaging linearity, dielectric layer interference, and limited fault detection and classification capabilities. Each of these problems will now be discussed in turn.
One limitation of prior art optical spot grid array implementations arises from the fact that inspecting with confocal imaging requires very tight focus control, which is very difficult to achieve at high scan rates with large numerical aperture short-wavelength optics. To overcome this problem, simultaneous multi-height confocal imaging is necessary. However, while taking several height-slice images sequentially, as described in the prior art, is compatible with a one frame review mode, it is not compatible with the continuous motion requirements of inspection systems.
Another limitation to large arrays in the prior art is the linearity requirement on the lens array, imaging optics and detector arrays. To obtain good results from a spot grid array system, close tolerances on the linearity of the optics is important—both for the microlens array and for the de-magnification optical elements. The optical spots must be located on an exactly rectilinear grid with very exact distances between the spots. Such extreme linearity is difficult and expensive to achieve.
Another limitation of prior art technology is the need to employ a coherent laser source to achieve sufficient power density for high-speed inspection. Many inspected substrates are covered by transparent or semitransparent dielectric layers, which cause interference phenomena between the surfaces of the dielectric layers. As the thickness of these layers varies across the wafer, the phase of the reflections of the coherent light from the top and bottom of the dielectric layer varies. Moreover, the interference can be either constructive or destructive. These interference phenomena cause a change in the reflected power despite the absence of defects or irregularities, limiting the accuracy of defect detection and thereby limiting the capability of the system to identify true defects.
A further limitation of prior art spot grid array techniques arises from the limited fault detection and classification ability resulting from the collection of light signals from a single angular section of an object. As a result, fault detection and analysis may require more than a single inspection, thus dramatically increasing the amount of data that needs to be processed and collected for reliable detection and classification of faults.
There exists a need for a low-cost, accurate, high-speed imaging system with a large FOV for reducing manufacturing costs and increasing production throughput.